Add Over-Searching in Search-Augmented Large Language Models (EACL 2026)

[ruoyuxie]

Hello! Thank you for maintaining this awesome deep-reasoning collection. I'd like to add our paper, which is directly relevant to the O1/R1 family and reasoning-with-search line of work covered here:

"Over-Searching in Search-Augmented Large Language Models" (EACL 2026) studies a key failure mode of reasoning models that invoke retrieval as a tool — they often over-search, calling retrieval even when it hurts response quality. We introduce Tokens Per Correctness (TPC) to measure the reasoning-efficiency vs. correctness trade-off, release the OverSearchQA benchmark, and study mitigation strategies. This complements existing entries here (Search-R1, R1-Searcher, ReSearch) by examining when not to search during reasoning.

• Paper: https://arxiv.org/abs/2601.05503
• Code: https://github.com/apple/ml-over-searching
• Venue: EACL 2026 (Main Conference)

Added a new 2026.01 section. Thanks!

[gemini-code-assist]

Code Review

This pull request adds a new paper entry for 'OverSearchQA' under a new '2026.01' section in the README.md. The reviewer suggested adding the repository link for the paper's open-source implementation to match the existing format.